blob: 744fa611cae06b26d89a04e915f0d7fadf37035b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/kernel/softirq.c
3 *
4 * Copyright (C) 1992 Linus Torvalds
5 *
Pavel Machekb10db7f2008-01-30 13:30:00 +01006 * Distribute under GPLv2.
7 *
8 * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Joe Perches40322762014-01-27 17:07:15 -080011#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
Paul Gortmaker9984de12011-05-23 14:51:41 -040013#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel_stat.h>
15#include <linux/interrupt.h>
16#include <linux/init.h>
17#include <linux/mm.h>
18#include <linux/notifier.h>
19#include <linux/percpu.h>
20#include <linux/cpu.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070021#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kthread.h>
23#include <linux/rcupdate.h>
Steven Rostedt7e49fcc2009-01-22 19:01:40 -050024#include <linux/ftrace.h>
Andrew Morton78eef012006-03-22 00:08:16 -080025#include <linux/smp.h>
Thomas Gleixner3e339b52012-07-16 10:42:37 +000026#include <linux/smpboot.h>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -080027#include <linux/tick.h>
Thomas Gleixnerd5326762014-03-19 11:19:52 +010028#include <linux/irq.h>
Heiko Carstensa0e39ed2009-04-29 13:51:39 +020029
30#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040031#include <trace/events/irq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*
34 - No shared variables, all the data are CPU local.
35 - If a softirq needs serialization, let it serialize itself
36 by its own spinlocks.
37 - Even if softirq is serialized, only local cpu is marked for
38 execution. Hence, we get something sort of weak cpu binding.
39 Though it is still not clear, will it result in better locality
40 or will not.
41
42 Examples:
43 - NET RX softirq. It is multithreaded and does not require
44 any global serialization.
45 - NET TX softirq. It kicks software netdevice queues, hence
46 it is logically serialized per device, but this serialization
47 is invisible to common code.
48 - Tasklets: serialized wrt itself.
49 */
50
51#ifndef __ARCH_IRQ_STAT
52irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
53EXPORT_SYMBOL(irq_stat);
54#endif
55
Alexey Dobriyan978b0112008-09-06 20:04:36 +020056static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Venkatesh Pallipadi4dd53d82010-12-21 17:09:00 -080058DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Joe Perchesce85b4f2014-01-27 17:07:16 -080060const char * const softirq_to_name[NR_SOFTIRQS] = {
Sagi Grimbergf660f602016-10-10 15:10:51 +030061 "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL",
Shaohua Li09223372011-06-14 13:26:25 +080062 "TASKLET", "SCHED", "HRTIMER", "RCU"
Jason Baron5d592b42009-03-12 14:33:36 -040063};
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
66 * we cannot loop indefinitely here to avoid userspace starvation,
67 * but we also don't want to introduce a worst case 1/HZ latency
68 * to the pending events, so lets the scheduler to balance
69 * the softirq load for us.
70 */
Thomas Gleixner676cb022009-07-20 23:33:49 +020071static void wakeup_softirqd(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 /* Interrupts are disabled: no need to stop preemption */
Christoph Lameter909ea962010-12-08 16:22:55 +010074 struct task_struct *tsk = __this_cpu_read(ksoftirqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 if (tsk && tsk->state != TASK_RUNNING)
77 wake_up_process(tsk);
78}
79
80/*
Eric Dumazet4cd13c22016-08-31 10:42:29 -070081 * If ksoftirqd is scheduled, we do not want to process pending softirqs
82 * right now. Let ksoftirqd handle this at its own rate, to get fairness.
83 */
84static bool ksoftirqd_running(void)
85{
86 struct task_struct *tsk = __this_cpu_read(ksoftirqd);
87
88 return tsk && (tsk->state == TASK_RUNNING);
89}
90
91/*
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -070092 * preempt_count and SOFTIRQ_OFFSET usage:
93 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
94 * softirq processing.
95 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
96 * on local_bh_disable or local_bh_enable.
97 * This lets us distinguish between whether we are currently processing
98 * softirq and whether we just have bh disabled.
99 */
100
101/*
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700102 * This one is for softirq.c-internal use,
103 * where hardirqs are disabled legitimately:
104 */
Tim Chen3c829c32006-07-30 03:04:02 -0700105#ifdef CONFIG_TRACE_IRQFLAGS
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100106void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700107{
108 unsigned long flags;
109
110 WARN_ON_ONCE(in_irq());
111
112 raw_local_irq_save(flags);
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500113 /*
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200114 * The preempt tracer hooks into preempt_count_add and will break
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500115 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
116 * is set and before current->softirq_enabled is cleared.
117 * We must manually increment preempt_count here and manually
118 * call the trace_preempt_off later.
119 */
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200120 __preempt_count_add(cnt);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700121 /*
122 * Were softirqs turned off above:
123 */
Peter Zijlstra9ea4c382013-11-19 16:13:38 +0100124 if (softirq_count() == (cnt & SOFTIRQ_MASK))
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700125 trace_softirqs_off(ip);
126 raw_local_irq_restore(flags);
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500127
Heiko Carstens0f1ba9a2015-01-07 10:04:41 +0100128 if (preempt_count() == cnt) {
129#ifdef CONFIG_DEBUG_PREEMPT
Sebastian Andrzej Siewiorf904f582016-02-26 14:54:56 +0100130 current->preempt_disable_ip = get_lock_parent_ip();
Heiko Carstens0f1ba9a2015-01-07 10:04:41 +0100131#endif
Sebastian Andrzej Siewiorf904f582016-02-26 14:54:56 +0100132 trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
Heiko Carstens0f1ba9a2015-01-07 10:04:41 +0100133 }
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700134}
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100135EXPORT_SYMBOL(__local_bh_disable_ip);
Tim Chen3c829c32006-07-30 03:04:02 -0700136#endif /* CONFIG_TRACE_IRQFLAGS */
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700137
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700138static void __local_bh_enable(unsigned int cnt)
139{
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700140 WARN_ON_ONCE(!irqs_disabled());
141
Peter Zijlstra9ea4c382013-11-19 16:13:38 +0100142 if (softirq_count() == (cnt & SOFTIRQ_MASK))
Davidlohr Buesod2e08472013-04-30 11:46:09 -0700143 trace_softirqs_on(_RET_IP_);
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200144 preempt_count_sub(cnt);
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700145}
146
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700147/*
148 * Special-case - softirqs can safely be enabled in
149 * cond_resched_softirq(), or by __do_softirq(),
150 * without processing still-pending softirqs:
151 */
152void _local_bh_enable(void)
153{
Frederic Weisbecker5d60d3e2013-09-24 04:11:35 +0200154 WARN_ON_ONCE(in_irq());
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700155 __local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700156}
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700157EXPORT_SYMBOL(_local_bh_enable);
158
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100159void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700160{
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200161 WARN_ON_ONCE(in_irq() || irqs_disabled());
Tim Chen3c829c32006-07-30 03:04:02 -0700162#ifdef CONFIG_TRACE_IRQFLAGS
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200163 local_irq_disable();
Tim Chen3c829c32006-07-30 03:04:02 -0700164#endif
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700165 /*
166 * Are softirqs going to be turned on now:
167 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700168 if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700169 trace_softirqs_on(ip);
170 /*
171 * Keep preemption disabled until we are done with
172 * softirq processing:
Joe Perchesce85b4f2014-01-27 17:07:16 -0800173 */
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100174 preempt_count_sub(cnt - 1);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700175
Frederic Weisbecker0bed6982013-09-05 16:14:00 +0200176 if (unlikely(!in_interrupt() && local_softirq_pending())) {
177 /*
178 * Run softirq if any pending. And do it in its own stack
179 * as we may be calling this deep in a task call stack already.
180 */
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700181 do_softirq();
Frederic Weisbecker0bed6982013-09-05 16:14:00 +0200182 }
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700183
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200184 preempt_count_dec();
Tim Chen3c829c32006-07-30 03:04:02 -0700185#ifdef CONFIG_TRACE_IRQFLAGS
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200186 local_irq_enable();
Tim Chen3c829c32006-07-30 03:04:02 -0700187#endif
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700188 preempt_check_resched();
189}
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100190EXPORT_SYMBOL(__local_bh_enable_ip);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700191
192/*
Ben Greear34376a52013-06-06 14:29:49 -0700193 * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
194 * but break the loop if need_resched() is set or after 2 ms.
195 * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
196 * certain cases, such as stop_machine(), jiffies may cease to
197 * increment and so we need the MAX_SOFTIRQ_RESTART limit as
198 * well to make sure we eventually return from this method.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 *
Eric Dumazetc10d7362013-01-10 15:26:34 -0800200 * These limits have been established via experimentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * The two things to balance is latency against fairness -
202 * we want to handle softirqs as soon as possible, but they
203 * should not be able to lock up the box.
204 */
Eric Dumazetc10d7362013-01-10 15:26:34 -0800205#define MAX_SOFTIRQ_TIME msecs_to_jiffies(2)
Ben Greear34376a52013-06-06 14:29:49 -0700206#define MAX_SOFTIRQ_RESTART 10
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100208#ifdef CONFIG_TRACE_IRQFLAGS
209/*
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100210 * When we run softirqs from irq_exit() and thus on the hardirq stack we need
211 * to keep the lockdep irq context tracking as tight as possible in order to
212 * not miss-qualify lock contexts and miss possible deadlocks.
213 */
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100214
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100215static inline bool lockdep_softirq_start(void)
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100216{
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100217 bool in_hardirq = false;
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100218
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100219 if (trace_hardirq_context(current)) {
220 in_hardirq = true;
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100221 trace_hardirq_exit();
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100222 }
223
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100224 lockdep_softirq_enter();
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100225
226 return in_hardirq;
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100227}
228
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100229static inline void lockdep_softirq_end(bool in_hardirq)
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100230{
231 lockdep_softirq_exit();
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100232
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100233 if (in_hardirq)
234 trace_hardirq_enter();
235}
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100236#else
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100237static inline bool lockdep_softirq_start(void) { return false; }
238static inline void lockdep_softirq_end(bool in_hardirq) { }
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100239#endif
240
Alexander Potapenkobe7635e2016-03-25 14:22:05 -0700241asmlinkage __visible void __softirq_entry __do_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Eric Dumazetc10d7362013-01-10 15:26:34 -0800243 unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
Mel Gorman907aed42012-07-31 16:44:07 -0700244 unsigned long old_flags = current->flags;
Ben Greear34376a52013-06-06 14:29:49 -0700245 int max_restart = MAX_SOFTIRQ_RESTART;
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100246 struct softirq_action *h;
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100247 bool in_hardirq;
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100248 __u32 pending;
Joe Perches2e702b92014-01-27 17:07:14 -0800249 int softirq_bit;
Mel Gorman907aed42012-07-31 16:44:07 -0700250
251 /*
252 * Mask out PF_MEMALLOC s current task context is borrowed for the
253 * softirq. A softirq handled such as network RX might set PF_MEMALLOC
254 * again if the socket is related to swap
255 */
256 current->flags &= ~PF_MEMALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 pending = local_softirq_pending();
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100259 account_irq_enter_time(current);
Paul Mackerras829035fd2006-07-03 00:25:40 -0700260
Peter Zijlstra0bd3a172013-11-19 16:13:38 +0100261 __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100262 in_hardirq = lockdep_softirq_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264restart:
265 /* Reset the pending bitmask before enabling irqs */
Andi Kleen3f744782005-09-12 18:49:24 +0200266 set_softirq_pending(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Andrew Mortonc70f5d62005-07-30 10:22:49 -0700268 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270 h = softirq_vec;
271
Joe Perches2e702b92014-01-27 17:07:14 -0800272 while ((softirq_bit = ffs(pending))) {
273 unsigned int vec_nr;
274 int prev_count;
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200275
Joe Perches2e702b92014-01-27 17:07:14 -0800276 h += softirq_bit - 1;
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200277
Joe Perches2e702b92014-01-27 17:07:14 -0800278 vec_nr = h - softirq_vec;
279 prev_count = preempt_count();
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200280
Joe Perches2e702b92014-01-27 17:07:14 -0800281 kstat_incr_softirqs_this_cpu(vec_nr);
282
283 trace_softirq_entry(vec_nr);
284 h->action(h);
285 trace_softirq_exit(vec_nr);
286 if (unlikely(prev_count != preempt_count())) {
Joe Perches40322762014-01-27 17:07:15 -0800287 pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
Joe Perches2e702b92014-01-27 17:07:14 -0800288 vec_nr, softirq_to_name[vec_nr], h->action,
289 prev_count, preempt_count());
290 preempt_count_set(prev_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
292 h++;
Joe Perches2e702b92014-01-27 17:07:14 -0800293 pending >>= softirq_bit;
294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Paul E. McKenney284a8c92014-08-14 16:38:46 -0700296 rcu_bh_qs();
Andrew Mortonc70f5d62005-07-30 10:22:49 -0700297 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
299 pending = local_softirq_pending();
Eric Dumazetc10d7362013-01-10 15:26:34 -0800300 if (pending) {
Ben Greear34376a52013-06-06 14:29:49 -0700301 if (time_before(jiffies, end) && !need_resched() &&
302 --max_restart)
Eric Dumazetc10d7362013-01-10 15:26:34 -0800303 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 wakeup_softirqd();
Eric Dumazetc10d7362013-01-10 15:26:34 -0800306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Frederic Weisbecker5c4853b2013-11-20 01:07:34 +0100308 lockdep_softirq_end(in_hardirq);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100309 account_irq_exit_time(current);
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700310 __local_bh_enable(SOFTIRQ_OFFSET);
Frederic Weisbecker5d60d3e2013-09-24 04:11:35 +0200311 WARN_ON_ONCE(in_interrupt());
Mel Gorman907aed42012-07-31 16:44:07 -0700312 tsk_restore_flags(current, old_flags, PF_MEMALLOC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
314
Andi Kleen722a9f92014-05-02 00:44:38 +0200315asmlinkage __visible void do_softirq(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 __u32 pending;
318 unsigned long flags;
319
320 if (in_interrupt())
321 return;
322
323 local_irq_save(flags);
324
325 pending = local_softirq_pending();
326
Eric Dumazet4cd13c22016-08-31 10:42:29 -0700327 if (pending && !ksoftirqd_running())
Frederic Weisbecker7d65f4a2013-09-05 15:49:45 +0200328 do_softirq_own_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 local_irq_restore(flags);
331}
332
Ingo Molnardde4b2b2007-02-16 01:27:45 -0800333/*
334 * Enter an interrupt context.
335 */
336void irq_enter(void)
337{
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100338 rcu_irq_enter();
Frederic Weisbecker0a8a2e72012-01-24 18:59:44 +0100339 if (is_idle_task(current) && !in_interrupt()) {
Venkatesh Pallipadid267f872010-10-04 17:03:23 -0700340 /*
341 * Prevent raise_softirq from needlessly waking up ksoftirqd
342 * here, as softirq will be serviced on return from interrupt.
343 */
344 local_bh_disable();
Frederic Weisbecker5acac1b2013-12-04 18:28:20 +0100345 tick_irq_enter();
Venkatesh Pallipadid267f872010-10-04 17:03:23 -0700346 _local_bh_enable();
347 }
348
349 __irq_enter();
Ingo Molnardde4b2b2007-02-16 01:27:45 -0800350}
351
Heiko Carstensb2a00172012-03-05 15:07:25 -0800352static inline void invoke_softirq(void)
353{
Eric Dumazet4cd13c22016-08-31 10:42:29 -0700354 if (ksoftirqd_running())
355 return;
356
Frederic Weisbeckerded79752013-09-24 00:50:25 +0200357 if (!force_irqthreads) {
Frederic Weisbeckercc1f0272013-09-24 17:17:47 +0200358#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
Frederic Weisbeckerded79752013-09-24 00:50:25 +0200359 /*
360 * We can safely execute softirq on the current stack if
361 * it is the irq stack, because it should be near empty
Frederic Weisbeckercc1f0272013-09-24 17:17:47 +0200362 * at this stage.
363 */
364 __do_softirq();
365#else
366 /*
367 * Otherwise, irq_exit() is called on the task stack that can
368 * be potentially deep already. So call softirq in its own stack
369 * to prevent from any overrun.
Frederic Weisbeckerded79752013-09-24 00:50:25 +0200370 */
Frederic Weisbeckerbe6e1012013-09-24 16:39:41 +0200371 do_softirq_own_stack();
Frederic Weisbeckercc1f0272013-09-24 17:17:47 +0200372#endif
Frederic Weisbeckerded79752013-09-24 00:50:25 +0200373 } else {
Thomas Gleixner8d32a302011-02-23 23:52:23 +0000374 wakeup_softirqd();
Frederic Weisbeckerded79752013-09-24 00:50:25 +0200375 }
Thomas Gleixner8d32a302011-02-23 23:52:23 +0000376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Frederic Weisbecker67826ea2013-04-20 17:43:13 +0200378static inline void tick_irq_exit(void)
379{
380#ifdef CONFIG_NO_HZ_COMMON
381 int cpu = smp_processor_id();
382
383 /* Make sure that timer wheel updates are propagated */
384 if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
385 if (!in_interrupt())
386 tick_nohz_irq_exit();
387 }
388#endif
389}
390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391/*
392 * Exit an interrupt context. Process softirqs if needed and possible:
393 */
394void irq_exit(void)
395{
Thomas Gleixner74eed012013-02-20 22:00:48 +0100396#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
Frederic Weisbecker4cd5d112013-02-28 20:00:43 +0100397 local_irq_disable();
Thomas Gleixner74eed012013-02-20 22:00:48 +0100398#else
399 WARN_ON_ONCE(!irqs_disabled());
400#endif
401
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100402 account_irq_exit_time(current);
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200403 preempt_count_sub(HARDIRQ_OFFSET);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 if (!in_interrupt() && local_softirq_pending())
405 invoke_softirq();
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -0800406
Frederic Weisbecker67826ea2013-04-20 17:43:13 +0200407 tick_irq_exit();
Frederic Weisbecker416eb332011-10-07 16:31:02 -0700408 rcu_irq_exit();
Peter Zijlstraf1a83e62013-11-19 16:42:47 +0100409 trace_hardirq_exit(); /* must be last! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
412/*
413 * This function must run with irqs disabled!
414 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800415inline void raise_softirq_irqoff(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
417 __raise_softirq_irqoff(nr);
418
419 /*
420 * If we're in an interrupt or softirq, we're done
421 * (this also catches softirq-disabled code). We will
422 * actually run the softirq once we return from
423 * the irq or softirq.
424 *
425 * Otherwise we wake up ksoftirqd to make sure we
426 * schedule the softirq soon.
427 */
428 if (!in_interrupt())
429 wakeup_softirqd();
430}
431
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800432void raise_softirq(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 unsigned long flags;
435
436 local_irq_save(flags);
437 raise_softirq_irqoff(nr);
438 local_irq_restore(flags);
439}
440
Steven Rostedtf0696862012-01-25 20:18:55 -0500441void __raise_softirq_irqoff(unsigned int nr)
442{
443 trace_softirq_raise(nr);
444 or_softirq_pending(1UL << nr);
445}
446
Carlos R. Mafra962cf362008-05-15 11:15:37 -0300447void open_softirq(int nr, void (*action)(struct softirq_action *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 softirq_vec[nr].action = action;
450}
451
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200452/*
453 * Tasklets
454 */
Joe Perchesce85b4f2014-01-27 17:07:16 -0800455struct tasklet_head {
Olof Johansson48f20a92008-03-04 15:23:25 -0800456 struct tasklet_struct *head;
457 struct tasklet_struct **tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458};
459
Vegard Nossum4620b492008-06-12 23:21:53 +0200460static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
461static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800463void __tasklet_schedule(struct tasklet_struct *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
465 unsigned long flags;
466
467 local_irq_save(flags);
Olof Johansson48f20a92008-03-04 15:23:25 -0800468 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100469 *__this_cpu_read(tasklet_vec.tail) = t;
470 __this_cpu_write(tasklet_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 raise_softirq_irqoff(TASKLET_SOFTIRQ);
472 local_irq_restore(flags);
473}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474EXPORT_SYMBOL(__tasklet_schedule);
475
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800476void __tasklet_hi_schedule(struct tasklet_struct *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
478 unsigned long flags;
479
480 local_irq_save(flags);
Olof Johansson48f20a92008-03-04 15:23:25 -0800481 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100482 *__this_cpu_read(tasklet_hi_vec.tail) = t;
483 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 raise_softirq_irqoff(HI_SOFTIRQ);
485 local_irq_restore(flags);
486}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487EXPORT_SYMBOL(__tasklet_hi_schedule);
488
Vegard Nossum7c692cb2008-05-21 22:53:13 +0200489void __tasklet_hi_schedule_first(struct tasklet_struct *t)
490{
491 BUG_ON(!irqs_disabled());
492
Christoph Lameter909ea962010-12-08 16:22:55 +0100493 t->next = __this_cpu_read(tasklet_hi_vec.head);
494 __this_cpu_write(tasklet_hi_vec.head, t);
Vegard Nossum7c692cb2008-05-21 22:53:13 +0200495 __raise_softirq_irqoff(HI_SOFTIRQ);
496}
Vegard Nossum7c692cb2008-05-21 22:53:13 +0200497EXPORT_SYMBOL(__tasklet_hi_schedule_first);
498
Emese Revfy0766f782016-06-20 20:42:34 +0200499static __latent_entropy void tasklet_action(struct softirq_action *a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 struct tasklet_struct *list;
502
503 local_irq_disable();
Christoph Lameter909ea962010-12-08 16:22:55 +0100504 list = __this_cpu_read(tasklet_vec.head);
505 __this_cpu_write(tasklet_vec.head, NULL);
Christoph Lameter22127e92014-08-17 12:30:25 -0500506 __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 local_irq_enable();
508
509 while (list) {
510 struct tasklet_struct *t = list;
511
512 list = list->next;
513
514 if (tasklet_trylock(t)) {
515 if (!atomic_read(&t->count)) {
Joe Perchesce85b4f2014-01-27 17:07:16 -0800516 if (!test_and_clear_bit(TASKLET_STATE_SCHED,
517 &t->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 BUG();
519 t->func(t->data);
520 tasklet_unlock(t);
521 continue;
522 }
523 tasklet_unlock(t);
524 }
525
526 local_irq_disable();
Olof Johansson48f20a92008-03-04 15:23:25 -0800527 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100528 *__this_cpu_read(tasklet_vec.tail) = t;
529 __this_cpu_write(tasklet_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 __raise_softirq_irqoff(TASKLET_SOFTIRQ);
531 local_irq_enable();
532 }
533}
534
Emese Revfy0766f782016-06-20 20:42:34 +0200535static __latent_entropy void tasklet_hi_action(struct softirq_action *a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 struct tasklet_struct *list;
538
539 local_irq_disable();
Christoph Lameter909ea962010-12-08 16:22:55 +0100540 list = __this_cpu_read(tasklet_hi_vec.head);
541 __this_cpu_write(tasklet_hi_vec.head, NULL);
Christoph Lameter22127e92014-08-17 12:30:25 -0500542 __this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 local_irq_enable();
544
545 while (list) {
546 struct tasklet_struct *t = list;
547
548 list = list->next;
549
550 if (tasklet_trylock(t)) {
551 if (!atomic_read(&t->count)) {
Joe Perchesce85b4f2014-01-27 17:07:16 -0800552 if (!test_and_clear_bit(TASKLET_STATE_SCHED,
553 &t->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 BUG();
555 t->func(t->data);
556 tasklet_unlock(t);
557 continue;
558 }
559 tasklet_unlock(t);
560 }
561
562 local_irq_disable();
Olof Johansson48f20a92008-03-04 15:23:25 -0800563 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100564 *__this_cpu_read(tasklet_hi_vec.tail) = t;
565 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 __raise_softirq_irqoff(HI_SOFTIRQ);
567 local_irq_enable();
568 }
569}
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571void tasklet_init(struct tasklet_struct *t,
572 void (*func)(unsigned long), unsigned long data)
573{
574 t->next = NULL;
575 t->state = 0;
576 atomic_set(&t->count, 0);
577 t->func = func;
578 t->data = data;
579}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580EXPORT_SYMBOL(tasklet_init);
581
582void tasklet_kill(struct tasklet_struct *t)
583{
584 if (in_interrupt())
Joe Perches40322762014-01-27 17:07:15 -0800585 pr_notice("Attempt to kill tasklet from interrupt\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
H Hartley Sweeten79d381c2009-04-16 19:30:18 -0400588 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 yield();
H Hartley Sweeten79d381c2009-04-16 19:30:18 -0400590 } while (test_bit(TASKLET_STATE_SCHED, &t->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
592 tasklet_unlock_wait(t);
593 clear_bit(TASKLET_STATE_SCHED, &t->state);
594}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595EXPORT_SYMBOL(tasklet_kill);
596
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200597/*
598 * tasklet_hrtimer
599 */
600
601/*
Peter Zijlstrab9c30322010-02-03 18:08:52 +0100602 * The trampoline is called when the hrtimer expires. It schedules a tasklet
603 * to run __tasklet_hrtimer_trampoline() which in turn will call the intended
604 * hrtimer callback, but from softirq context.
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200605 */
606static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
607{
608 struct tasklet_hrtimer *ttimer =
609 container_of(timer, struct tasklet_hrtimer, timer);
610
Peter Zijlstrab9c30322010-02-03 18:08:52 +0100611 tasklet_hi_schedule(&ttimer->tasklet);
612 return HRTIMER_NORESTART;
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200613}
614
615/*
616 * Helper function which calls the hrtimer callback from
617 * tasklet/softirq context
618 */
619static void __tasklet_hrtimer_trampoline(unsigned long data)
620{
621 struct tasklet_hrtimer *ttimer = (void *)data;
622 enum hrtimer_restart restart;
623
624 restart = ttimer->function(&ttimer->timer);
625 if (restart != HRTIMER_NORESTART)
626 hrtimer_restart(&ttimer->timer);
627}
628
629/**
630 * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
631 * @ttimer: tasklet_hrtimer which is initialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300632 * @function: hrtimer callback function which gets called from softirq context
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200633 * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
634 * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
635 */
636void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
637 enum hrtimer_restart (*function)(struct hrtimer *),
638 clockid_t which_clock, enum hrtimer_mode mode)
639{
640 hrtimer_init(&ttimer->timer, which_clock, mode);
641 ttimer->timer.function = __hrtimer_tasklet_trampoline;
642 tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
643 (unsigned long)ttimer);
644 ttimer->function = function;
645}
646EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648void __init softirq_init(void)
649{
Olof Johansson48f20a92008-03-04 15:23:25 -0800650 int cpu;
651
652 for_each_possible_cpu(cpu) {
653 per_cpu(tasklet_vec, cpu).tail =
654 &per_cpu(tasklet_vec, cpu).head;
655 per_cpu(tasklet_hi_vec, cpu).tail =
656 &per_cpu(tasklet_hi_vec, cpu).head;
657 }
658
Carlos R. Mafra962cf362008-05-15 11:15:37 -0300659 open_softirq(TASKLET_SOFTIRQ, tasklet_action);
660 open_softirq(HI_SOFTIRQ, tasklet_hi_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661}
662
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000663static int ksoftirqd_should_run(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000665 return local_softirq_pending();
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000668static void run_ksoftirqd(unsigned int cpu)
669{
670 local_irq_disable();
671 if (local_softirq_pending()) {
Frederic Weisbecker0bed6982013-09-05 16:14:00 +0200672 /*
673 * We can safely run softirq on inline stack, as we are not deep
674 * in the task stack here.
675 */
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000676 __do_softirq();
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000677 local_irq_enable();
Paul E. McKenney60479672015-01-14 13:20:26 -0800678 cond_resched_rcu_qs();
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000679 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 }
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000681 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
684#ifdef CONFIG_HOTPLUG_CPU
685/*
686 * tasklet_kill_immediate is called to remove a tasklet which can already be
687 * scheduled for execution on @cpu.
688 *
689 * Unlike tasklet_kill, this function removes the tasklet
690 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
691 *
692 * When this function is called, @cpu must be in the CPU_DEAD state.
693 */
694void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
695{
696 struct tasklet_struct **i;
697
698 BUG_ON(cpu_online(cpu));
699 BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
700
701 if (!test_bit(TASKLET_STATE_SCHED, &t->state))
702 return;
703
704 /* CPU is dead, so no lock needed. */
Olof Johansson48f20a92008-03-04 15:23:25 -0800705 for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 if (*i == t) {
707 *i = t->next;
Olof Johansson48f20a92008-03-04 15:23:25 -0800708 /* If this was the tail element, move the tail ptr */
709 if (*i == NULL)
710 per_cpu(tasklet_vec, cpu).tail = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return;
712 }
713 }
714 BUG();
715}
716
Sebastian Andrzej Siewiorc4544db2016-08-18 14:57:21 +0200717static int takeover_tasklets(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 /* CPU is dead, so no lock needed. */
720 local_irq_disable();
721
722 /* Find end, append list for that CPU. */
Christian Borntraegere5e41722008-05-01 04:34:23 -0700723 if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100724 *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
725 this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
Christian Borntraegere5e41722008-05-01 04:34:23 -0700726 per_cpu(tasklet_vec, cpu).head = NULL;
727 per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 raise_softirq_irqoff(TASKLET_SOFTIRQ);
730
Christian Borntraegere5e41722008-05-01 04:34:23 -0700731 if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100732 *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
733 __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
Christian Borntraegere5e41722008-05-01 04:34:23 -0700734 per_cpu(tasklet_hi_vec, cpu).head = NULL;
735 per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 raise_softirq_irqoff(HI_SOFTIRQ);
738
739 local_irq_enable();
Sebastian Andrzej Siewiorc4544db2016-08-18 14:57:21 +0200740 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
Sebastian Andrzej Siewiorc4544db2016-08-18 14:57:21 +0200742#else
743#define takeover_tasklets NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744#endif /* CONFIG_HOTPLUG_CPU */
745
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000746static struct smp_hotplug_thread softirq_threads = {
747 .store = &ksoftirqd,
748 .thread_should_run = ksoftirqd_should_run,
749 .thread_fn = run_ksoftirqd,
750 .thread_comm = "ksoftirqd/%u",
751};
752
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700753static __init int spawn_ksoftirqd(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
Sebastian Andrzej Siewiorc4544db2016-08-18 14:57:21 +0200755 cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL,
756 takeover_tasklets);
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000757 BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return 0;
760}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700761early_initcall(spawn_ksoftirqd);
Andrew Morton78eef012006-03-22 00:08:16 -0800762
Yinghai Lu43a25632008-12-28 16:01:13 -0800763/*
764 * [ These __weak aliases are kept in a separate compilation unit, so that
765 * GCC does not inline them incorrectly. ]
766 */
767
768int __init __weak early_irq_init(void)
769{
770 return 0;
771}
772
Yinghai Lu4a046d12009-01-12 17:39:24 -0800773int __init __weak arch_probe_nr_irqs(void)
774{
Thomas Gleixnerb683de22010-09-27 20:55:03 +0200775 return NR_IRQS_LEGACY;
Yinghai Lu4a046d12009-01-12 17:39:24 -0800776}
777
Yinghai Lu43a25632008-12-28 16:01:13 -0800778int __init __weak arch_early_irq_init(void)
779{
780 return 0;
781}
Thomas Gleixner62a08ae2014-04-24 09:50:53 +0200782
783unsigned int __weak arch_dynirq_lower_bound(unsigned int from)
784{
785 return from;
786}